Skip to main content
Upcoming Events:
Tools for improving process monitoring and control in primary aluminum smelting

Tools for improving process monitoring and control in primary aluminum smelting

Date & Time:
   Add All to Calendar

BSB 119 

Carl Duchesne, Universite Laval


Latent Variable Methods in Primary Aluminum Smelting

As new process analytical technologies (PAT) are developed and data infrastructures improved, the data collected from industrial processes are not only available in larger amounts, but are also more diversified and complex. Since their introduction in process engineering in the 1990’s, latent variable methods such as Principal Component Analysis (PCA), Projection to Latent Structures (PLS) and their variants, were successfully used for process analysis, monitoring and product quality control, and are still very useful today. To illustrate this, an overview of the research made in the last 10 years in the field of primary aluminum smelting in collaboration with Alcoa Corporation is proposed. The currently decreasing quality and increasing variability of raw materials combined with the lack of analytical sensors are important issues in aluminum production. Latent variable methods were used to develop several imaging and acousto-ultrasonic sensors for quality control of the carbon anodes consumed by the Hall-Héroult aluminum reduction cells. In the near future, a new anode tracking technology will allow data fusion between the anode manufacturing and the aluminum reduction plants, and enable plant-wide data analyses to be performed with an unprecedented level of details. The sequential pathway PLS algorithm developed recently will help cope with the enormous amount of data. For instance, it will be used to establish multivariate specification regions for incoming raw material properties in order to maintain high energy efficiency and reduce material overconsumption. After a brief introduction to latent variable methods, the talk will focus on presenting some of the results obtained so far, and will conclude with a discussion about future research directions and Big Data challenges.